From owner-freebsd-isp Fri Jul 11 19:57:54 1997 Return-Path: Received: (from root@localhost) by hub.freebsd.org (8.8.5/8.8.5) id TAA15882 for isp-outgoing; Fri, 11 Jul 1997 19:57:54 -0700 (PDT) Received: from mail.webspan.net (root@mail.webspan.net [206.154.70.7]) by hub.freebsd.org (8.8.5/8.8.5) with ESMTP id TAA15874 for ; Fri, 11 Jul 1997 19:57:51 -0700 (PDT) Received: from orion.webspan.net (orion.webspan.net [206.154.70.5]) by mail.webspan.net (WEBSPAN/970608) with ESMTP id WAA19943; Fri, 11 Jul 1997 22:57:50 -0400 (EDT) Received: from orion.webspan.net (localhost [127.0.0.1]) by orion.webspan.net (WEBSPAN/970608) with ESMTP id WAA02394; Fri, 11 Jul 1997 22:57:49 -0400 (EDT) To: John-Mark Gurney cc: David Lowe , freebsd-isp@FreeBSD.ORG From: "Gary Palmer" Subject: Re: INN on an async-mounted spool? In-reply-to: Your message of "Fri, 11 Jul 1997 04:05:06 PDT." <19970711040506.04604@hydrogen.nike.efn.org> Date: Fri, 11 Jul 1997 22:57:49 -0400 Message-ID: <2392.868676269@orion.webspan.net> Sender: owner-isp@FreeBSD.ORG X-Loop: FreeBSD.org Precedence: bulk John-Mark Gurney wrote in message ID <19970711040506.04604@hydrogen.nike.efn.org>: > Joe Greco runs a number of news machines... if I remeber correctly he > found the increase the sync interval (kern.update) to something like > 300 (5 minutes) will give you near the same performance as mounting > the fs's async, but give you the stability of sync mount fs... Really? Thats surprising. Async mounts don't flush out file metadata until sync() is called (from memory). Increasing the kern.update time on a non-async mount just decreases the frequency that file buffers are flushed. I'd have thought the no-metadata updates would have been a big win for a news server, especially when you are writing out >>8 articles/sec. noatime is an even bigger win (for reader boxes) > I would highly recommend anybody thinking about setting up a news > server to search the mailing lists for Greco and news to read his > postings on the subject... Indeed Joe has many pearls of wisdom to impart on the subject. :) > you also might want to look at using cdd to give your spools multiple > disks to work on... the more spindles the better.. but make sure that > the strip size is 32megs (or whatever the size of a cylinder group is) > to get the best performance... Just a note here: I'm not *100%* sure that a 32Mb stripe size is right. While on the face of it Joe's arguments in support of it make sense, I have yet to see any figures to support his assertions. I agree 100% that small stripes, as used by hardware controllers (e.g. 4 or 8k) are useless in the face of a full feed (unless you happen to stripe more drives together than most people see in a year). I just think 32Mb is going a bit too far in the other direction. If you are looking for a pure feed box, striping is no longer a win. You need to go to a radically different filesystem layout than what inn uses by default these days. e.g. the cyclical filesystem, or various others I've heard of which use the system clock as a base for pathnames to increase the localization of reference. Gary -- Gary Palmer FreeBSD Core Team Member FreeBSD: Turning PC's into workstations. See http://www.FreeBSD.ORG/ for info